Forty minutes is not long enough

The median sit in this record was forty minutes. The range ran from twenty to ninety, and the total came to about a hundred and forty-four hours across two hundred and six places.

Conversation between strangers is rare and arrives in bursts. Forty minutes catches a fraction of it. When observations span many locations, Monitask analytics for operational decisions can add time and project context without turning a single activity signal into a performance verdict.

What the field actually measures

Whether, during one sit of a particular length, on a particular day, at a particular hour, an exchange occurred in the room between two people who had not arrived together. Readers comparing this walk with broader public-space evidence can also see Walk Friendly Communities.

Not whether the room produces conversation. Not how often. Not whether it did yesterday.

Forty-seven is the number of places where it happened while somebody was watching. The number where it happens is higher.

How much higher

Unknown, and the record declines to estimate it.

What can be said is the shape of the error. It is one-directional: a longer sit can only move places from the silent column to the speaking one, never the reverse. So forty-seven is a floor and a hundred and forty-one is a ceiling.

The eighteen unobservable places widen the honest range at the top: if all of them had produced a conversation, the record's rate would be thirty-two per cent rather than twenty-three.

So the true figure for this walk, with unlimited watching, is somewhere above twenty-three per cent and the record cannot bound it above.

What survives the uncertainty

The order of the kinds.

A museum watched for forty minutes produced nothing in eight of eight. A pub watched for forty minutes produced something in seventeen of thirty-one. Extending both sits would raise both figures, and would have to raise the museum's by an implausible amount to close a forty-point gap.

So comparisons between kinds survive the window. Absolute rates do not, and no page on this site uses an absolute conversation rate for anything except to say it is a floor.

The hour, which is worse

A hundred and sixty-four of the two hundred and six visits were between ten and five on a weekday.

That is the quiet version of nearly every kind in the record, and the bias is not even. A library at eleven on a Tuesday is close to a library at any time. A pub at three on a Tuesday is not a pub.

If the hour bias ran uniformly it would depress every figure equally and leave the comparisons intact. It does not: it depresses the sociable kinds hardest, which means the gap between the pub and the museum is understated here rather than overstated.

That is the one place in this record where a known bias works against the finding rather than for it.

Why not sit longer

Because two hundred and six places at two hours each is four hundred hours, and the walk was eleven weeks.

A trade was made, explicitly, before the walk: breadth over depth. Two hundred and six places at forty minutes, rather than seventy places at two hours. The first gives twelve kinds with enough instances to compare; the second gives a better conversation figure and nothing to compare it with.

Given what the walk found — that the terms predict nothing and the kind predicts everything — the trade was probably right, and it is the reason the conversation field is the weakest on the site.

What a longer sit would and would not change

what moves              the absolute rates, all upward
what does not move      the order of the kinds

A museum watched for forty minutes produced nothing in eight of eight. A pub watched for forty minutes produced something in seventeen of thirty-one. Extending both would raise both, and would have to raise the museum's by an implausible amount to close a forty-point gap.

So every comparison between kinds on this site survives the window, and no absolute conversation figure does. No page uses one except to say it is a floor.

The arithmetic of the floor

seen to speak                          47    23%
seen not to speak                     141    68%
could not be seen                      18     9%

if every unseen place had spoken       65    32%
if none had                            47    23%

The honest range for the record's headline is twenty-three to thirty-two per cent, and twenty-three is printed because it is what was seen.

Above thirty-two there is no bound at all, because a longer sit in any of the hundred and forty-one silent places could move it.

The trade that was made

Two hundred and six places at forty minutes, or seventy places at two hours. The walk took the first.

The second gives a better conversation figure and nothing to compare it with. The first gives twelve kinds with enough instances to set against each other, which is where every result in this record came from.

Given that the terms turned out to predict nothing and the kind turned out to predict everything, the trade was probably right — and it is the reason the conversation field is the weakest on the site.

What would fix it

Three sits per place at different hours, including one evening and one weekend. Six hundred hours, or a third of the places.

Or a different instrument entirely: stay in one room all day and count exchanges rather than recording whether one happened. That measures the room properly and tells you nothing about a city.

The page on what a second walk would settle lists both, and the honest answer is that they are two different studies and this record is neither.